In our project, we used Yelp dataset to create visualizations and explore bar business in Las Vegas. We kept data only in Las Vegas, and used the datasets of “reviews”, “business” and user data. We mainly did exploratory data analysis, text analysis, geospatial analysis and social network analysis.

We spotted various star rating bars and provided interactive maps of the city and neighborhood of Las Vegas. For specific business, we did a word cloud of reviews of the bars, top 10 common words of the bars, sentiment analysis using “affin”, display of top positive reviews and bottom negative reviews of specific bars. For social network analysis, we extracted the top 10% elite users of Yelp, and visualized the networks of the users and bar business. Moreover, we also run a regression model to explore the relationship between bar ratings and review counts.

Data https://www.kaggle.com/datasets/yelp-dataset/yelp-dataset

#library(tidyverse) 
library(stringr)
library(wordcloud) 
library(tidytext) 
library(DT) 
library(dplyr)
library(leaflet)

Part 1: Map Visualization for Vegas Bars

library(readr)
library(ggplot2)
library(ggthemes)
library(maps)
library(dplyr)
library("DT")
library(stringr)
library("leaflet")
library(RColorBrewer)
library("data.table")
library(ggmap)
register_google(key = "AIzaSyBlZS1dMHmz5qAZww5VQRBBSlJa3VFfyVg", write = TRUE)

Las Vegas Map

map_lv <- get_map("Las Vegas",
                            zoom = 12,
                            source = "stamen",
                            maptype = "toner-background")
ggmap(map_lv)

Below is the map that shows bar location

g_location <- ggmap(map_lv) + theme_map()
g_location + geom_point(data=bar, aes(x=longitude,y=latitude),
                    size=0.3, alpha=0.3, color="blue")

Highlighting the bars in hot spot

g_density <- g_location + geom_density2d(aes(x=longitude,y=latitude), 
  data=bar, color="green", size=1, bins=12) +
  stat_density2d(aes(x=longitude,y=latitude,
    fill = ..level.., alpha = ..level..),
    data=bar, geom = 'polygon', bins=12) +
  scale_fill_gradient2(low = "green", mid="yellow", high = "red") +
  scale_alpha(range = c(0.00, 0.5)) 
g_density

Visualize the neighborhood each bar belongs to:

#add legend of stars
lvbar_map_neighborhood <- 
  leaflet(bar) %>%
  addTiles() %>%    # Add OpenStreetMap map tiles
  addCircles(lng = ~longitude, lat = ~latitude)

pal = colorFactor("Set1", domain = bar$neighborhood) # Grab a palette
color_neighborhood = pal(bar$neighborhood)

lvbar_map_neighborhood %>% addCircles(color=color_neighborhood) %>%
  addLegend(pal = pal, values = ~bar$neighborhood, title = "Neighborhood")
content <- paste("Name:",bar$name,"<br/>",
                 "Address:",bar$address,"<br/>",
                 "Stars:",bar$stars,"<br/>",
                 "Neighborhood:", bar$neighborhood,"<br/>")

pal = colorFactor("YlOrRd", domain = bar$stars) # Grab a palette
color_stars = pal(bar$stars)

lvbar_map_neighborhood %>% addCircles(color=color_stars, popup = content) %>%
  addLegend(pal = pal, values = ~bar$stars, title = "Stars")

Part2: text analysis

In this section, we want to use NLP techniques to draw insights from yelp reviews regarding bars in Las Vegas.

First, we visualizaed the most Popular categories regarding bars(some businesses belongs to various business categories, such as bar, nightlife, restaurants..)

fillColor = "#FFA07A"
fillColor2 = "#F1C40F"

categories = str_split(bar$categories,";")
categories = as.data.frame(unlist(categories))
colnames(categories) = c("Name")

categories %>%
  group_by(Name) %>%
  summarise(Count = n()) %>%
  arrange(desc(Count)) %>%
  ungroup() %>%
  mutate(Name = reorder(Name,Count)) %>%
  head(10) %>%
  
  
  ggplot(aes(x = Name,y = Count)) +
  geom_bar(stat='identity',colour="white", fill =fillColor2) +
  geom_text(aes(x = Name, y = 1, label = paste0("(",Count,")",sep="")),
            hjust=0, vjust=.5, size = 4, colour = 'black',
            fontface = 'bold') +
  labs(x = 'Name of Category', y = 'Count', 
       title = 'Top 10 Categories regarding bars') +
  coord_flip() + 
  theme_bw()

It seems that the top 10 categories regarding bars are nightlife,bars,restaurants, lounges, American,sport bars,food, pubs and cocktail bars.

Then, it’s curious to see which bars has the most number of five Star reviews, which indicate successful bar business:

stars_5 <- barreview %>%
  filter(stars ==5) %>%
  group_by(business_id) %>%
  select(business_id,stars,text) %>%
  summarise(Count = n()) %>%
  arrange(desc(Count)) %>%
  ungroup() 


five = merge(stars_5, bar, by= "business_id")
five2 <- five %>%
  filter(stars ==5) %>%
  filter(is_open==1)

fivestar <- five2 %>%
  arrange(stars) %>%
  head(10)
fillColor2 = "#F1C40F"

fivestar %>%
  mutate(name = reorder(name,Count)) %>%
  ggplot(aes(x = name,y = Count)) +
  geom_bar(stat='identity',colour="white", fill = fillColor2) +
  geom_text(aes(x = name, y = 1, label = paste0("(",Count,")",sep="")),
            hjust=0, vjust=.5, size = 2, colour = 'black',
            fontface = 'bold') +
  labs(x = 'Name of the Bars', 
       y = 'Count', 
       title = 'Name of the bars and Count') +
  coord_flip() +
  theme_bw()

Here we found the bar with the greatest number of 5 star reviews is J Karaoke Bar, and here we display more detail information about this bar:

J_karaoke = bar %>% filter(business_id == "3pSUr_cdrphurO6m1HMP9A") %>%
  select(name,neighborhood,city,state,postal_code,categories)

datatable(head(J_karaoke), style="bootstrap", class="table-condensed", options = list(dom = 'tp',scrollX = TRUE))

It’s interesting to see a wordcloud that contains some common words of reviews on “J Karaoke Bar”:

createWordCloud = function(train)
{
  train %>%
  unnest_tokens(word, text) %>%
  filter(!word %in% stop_words$word) %>%
  count(word,sort = TRUE) %>%
  ungroup()  %>%
  head(30) %>%
  
  with(wordcloud(word, n, max.words = 30,colors=brewer.pal(8, "Dark2")))
}

createWordCloud(barreview %>%
  filter(business_id == "3pSUr_cdrphurO6m1HMP9A"))

From the wordcloud, we can derive the ingisht that people praise the atmosphere, music, cleaning environment, services and food(especially chicken) in this bar, and indicates that they spend happy and comfortable time in this J Karaoke bar.

Similarly, there are the bars with most number of one star reviews, and we still want to explore which bar has the most number of one star review:

#library()
stars_1 <- barreview %>%
  filter(stars ==1|stars==1.5) %>%
  group_by(business_id) %>%
  select(business_id,stars,text) %>%
  summarise(Count = n()) %>%
  arrange(desc(Count)) %>%
  ungroup() 


one = merge(stars_1, bar, by= "business_id")
one2 <- one %>%
  filter(stars ==1|stars==1.5) %>%
  filter(is_open==1)

onestar <- one2 %>%
  arrange(stars) %>%
  head(10)

fillColor2 = "#F1C40F"

onestar %>%
  mutate(name = reorder(name,Count)) %>%
  ggplot(aes(x = name,y = Count)) +
  geom_bar(stat='identity',colour="white", fill = fillColor2) +
  geom_text(aes(x = name, y = 1, label = paste0("(",Count,")",sep="")),
            hjust=0, vjust=.5, size = 2, colour = 'black',
            fontface = 'bold') +
  labs(x = 'Name of the Bars', 
       y = 'Count', 
       title = 'Name of the bars and Count') +
  coord_flip() +
  theme_bw()

Surprisingly, the bar named “Triumph property management” only has one star rating, and there are 12 reviews on that bar.

So we are interested to see the common words of reviews on “Triumph property management”:

createWordCloud = function(train)
{
  train %>%
  unnest_tokens(word, text) %>%
  filter(!word %in% stop_words$word) %>%
  count(word,sort = TRUE) %>%
  ungroup()  %>%
  head(30) %>%
  
  with(wordcloud(word, n, max.words = 30,colors=brewer.pal(8, "Dark2")))
}

createWordCloud(barreview %>%
  filter(business_id == "Zh6fxrqsKqdSVmTK3roxBQ"))

People in their reviews complain about the house/environment of the bar seemingly.

Let’s create a datatable to see some information regarding “Triumph property management”:

Triumph = bar %>% filter(business_id == "Zh6fxrqsKqdSVmTK3roxBQ") %>%
  select(name,neighborhood,city,state,postal_code,categories)

datatable(head(Triumph), style="bootstrap", class="table-condensed", options = list(dom = 'tp',scrollX = TRUE))
goodbar <- barreview %>%
  filter(stars == 5) %>%
  group_by(business_id) %>%
  ungroup()

star_five <- merge(goodbar,bar,by = "business_id")

badbar <- barreview %>%
  filter(stars == 1) %>%
  group_by(business_id) %>%
  ungroup()

star_one <- merge(badbar,bar,by = "business_id")
library(tm)
df_five = data.frame(doc_id = star_five$business_id, text = star_five$text,stringsAsFactors = F)
star_five2 <- DataframeSource(df_five)
star_five2 <- VCorpus(star_five2)

df_one = data.frame(doc_id = star_one$business_id, text = star_one$text,stringsAsFactors = F)
star_one2 <- DataframeSource(df_one)
star_one2 <- VCorpus(star_one2)
#Remove unnecessary words(stop words), synatx, punctuation, numbers, white space etc.
library(stringr)
remove_nonalphanum <- function(x){str_replace_all(x, "[^[:alnum:]]", " ")}
remove_brandnames <- function(x){str_replace_all(x, "\\b[A-Z]+\\b", " ")}

clean_corpus <- function(corpus){
  corpus <- tm_map(corpus, removePunctuation)
  corpus <- tm_map(corpus, content_transformer(remove_nonalphanum))
  corpus <- tm_map(corpus, content_transformer(remove_brandnames))
  corpus <- tm_map(corpus, content_transformer(tolower))
  corpus <- tm_map(corpus, removeWords, c(stopwords("en")))
  corpus <- tm_map(corpus, removeNumbers)
  corpus <- tm_map(corpus, stripWhitespace)
  return(corpus)
}
#cleaning two datasets
star_five_clean <- clean_corpus(star_five2)
star_one_clean <- clean_corpus(star_one2)
#create a document-term-matrix:
library(tm)
#create the dtm from the corpus
corpus_five_dtm <- DocumentTermMatrix(star_five_clean)
corpus_one_dtm <- DocumentTermMatrix(star_one_clean)
term_frequency_DT_five <- corpus_five_tdidf
term_frequency_DT_one <- corpus_one_tdidf

library(wordcloud)

After getting to know what common words in the reivews of the two top and bottom bars, we want to explore the general common words of reviews for five star and one star bar:

Here we create a wordcloud to show the frequent words of five stars bars: We can see people generally have great words to describe their experience of the five star bars:

Words like great, cool, amazing, love,time and good appear frequently, which indicates people enjoy their time getting infused of alcohol, and they really love the bar.

#Set seed
set.seed(2103)

#create a wordcloud to show the frequent words of five stars bars
wordcloud(term_frequency_DT_five$term, term_frequency_DT_five$tf,
          max.words=50, colors=brewer.pal(8, "Dark2"))

As for one star bars, words like awful, waste, bad, sucks,lounge, servers appear frequently.

#create a wordcloud to show the frequent words of one stars bars
wordcloud(term_frequency_DT_one$term, term_frequency_DT_one$tf,
          max.words=50, colors=brewer.pal(8, "Dark2"))

Here we generate a pyramid plot to have a more intuitive comparison of the words between five-stars and one-stars bars differ in word frequency:

#combine corpus of the most successful and unsuccessful projects
#select top 20 words
corpus_five_dt$bestworst <- "Top"
corpus_one_dt$bestworst <- "Bottom"
corpus_top_bottom_dt <- rbind(corpus_five_dt,corpus_one_dt)


corpus_top_bottom_count <- corpus_top_bottom_dt %>%
  group_by(term) %>%
  summarize(total_word = sum(count)) %>%
  arrange(desc(total_word)) %>%
  head(20)

Given that one star bars are rare, so the review/words of that bar are fewer, we can see the from the pyramid plot that the word counts of one star bars are much less than that of five stars.But generally,words like food, place, great, food, services that people mention a lot.

pyramid = left_join(corpus_top_bottom_dt, corpus_top_bottom_count, by='term')

pyramid <- pyramid %>%
  filter(!is.na(total_word)) %>%
  group_by(bestworst) %>%
  mutate(count_plot = ifelse(bestworst == 'Bottom', count*(-1), count))
ggplot(pyramid, aes(x = reorder(term, total_word),
                  y = count_plot, fill = bestworst)) +
  geom_bar(data = filter(pyramid, bestworst == "Top"), stat = "identity") +
  geom_bar(data = filter(pyramid, bestworst == "Bottom"), stat = "identity") +
  scale_fill_brewer(palette = "Set1", direction=-1) +
  coord_flip() +
  scale_y_continuous(breaks = seq(-50,50,25)) +
  scale_fill_discrete(name = 'bars star rating', labels=c('one star', 'five star')) +
  ylab("") +
  ggthemes::theme_tufte() + 
  labs(
    x = 'Top 20 Words',
    y= 'Count',
    title = 'Pyramid Plot of Top 20 Words, for one star bars and five star bars'
  )

Sentiment analysis of reviews:

We will continue to explore the reviews of the two typical bars: first, we compare the positive and negative words in the reviews of J Karaoke Bar:

library(tidytext)
library(tidyverse)

positiveWordsBarGraph <- function(SC) {
  contributions <- SC %>%
    unnest_tokens(word, text) %>%
    count(word,sort = TRUE) %>%
    ungroup() %>%
    
    inner_join(get_sentiments("afinn"), by = "word") %>%
    group_by(word) %>%
    summarize(occurences = n(),
              contribution = sum(value))

  
  contributions %>%
    top_n(20, abs(contribution)) %>%
    mutate(word = reorder(word, contribution)) %>%
    head(20) %>%
    ggplot(aes(word, contribution, fill = contribution > 0)) +
    geom_col(show.legend = FALSE) +
    coord_flip() + theme_bw()
}

positiveWordsBarGraph(barreview %>%
                       filter(business_id == "3pSUr_cdrphurO6m1HMP9A"))

Display top 10 most positive reviews for 5 star bars:

display_pos_sentiments <- function(sentiment_lines,review_text)
{
  pos_sentiment_lines = sentiment_lines %>%
  arrange(desc(sentiment))  %>%
  top_n(10, sentiment) %>%
  inner_join(review_text, by = "user_id") %>%
  select(date,sentiment,text) 
  
datatable(pos_sentiment_lines, style="bootstrap", class="table-condensed", options = list(dom = 'tp',scrollX = TRUE))

}

display_pos_sentiments(sentiment_lines,J_Karaoke_reviews)

Then, we compare the positive and negative words in the reviews of the positive and negative words in the reviews of Triumph property management:

positiveWordsBarGraph <- function(SC) {
  contributions <- SC %>%
    unnest_tokens(word, text) %>%
    count(word,sort = TRUE) %>%
    ungroup() %>%
    
    inner_join(get_sentiments("afinn"), by = "word") %>%
    group_by(word) %>%
    summarize(occurences = n(),
              contribution = sum(value))

  
  contributions %>%
    top_n(20, abs(contribution)) %>%
    mutate(word = reorder(word, contribution)) %>%
    head(20) %>%
    ggplot(aes(word, contribution, fill = contribution > 0)) +
    geom_col(show.legend = FALSE) +
    coord_flip() + theme_bw()
}

positiveWordsBarGraph(barreview %>%
                       filter(business_id == "Zh6fxrqsKqdSVmTK3roxBQ"))

Triumph_reviews = barreview %>%
  filter(business_id == "Zh6fxrqsKqdSVmTK3roxBQ")

calculate_sentiment <- function(review)
{
  sentiment_lines  =  review %>%
                  unnest_tokens(word, text) %>%
                  inner_join(get_sentiments("afinn"), by = "word") %>%
                  group_by(user_id) %>%
                  summarize(sentiment = mean(value),words = n()) %>%
                  ungroup() %>%
                  filter(words >= 10) 

  return(sentiment_lines)
  
}


sentiment_lines = calculate_sentiment(Triumph_reviews)

Display top 10 most negative reviews for Triumph property management:

display_neg_sentiments <- function(sentiment_lines,review_text)
{
  neg_sentiment_lines = sentiment_lines %>%
  arrange(desc(sentiment))  %>%
  top_n(-10, sentiment) %>%
  inner_join(review_text, by = "user_id") %>%
  select(date,sentiment,text) 
  
datatable(neg_sentiment_lines, style="bootstrap", class="table-condensed", options = list(dom = 'tp',scrollX = TRUE))
}

display_neg_sentiments(sentiment_lines,Triumph_reviews)

Here is the visualization of the geographical location of the top 10 five star bars(blue dot) and bottom 1 star or 1.5 star bars(yellow dot):

LasvegasCoords = bar %>% filter(city == "Las Vegas")
center_lon = median(LasvegasCoords$longitude,na.rm = TRUE)
center_lat = median(LasvegasCoords$latitude,na.rm = TRUE)

map <- leaflet(rbind(fivestar,onestar)) %>%
  addProviderTiles("Esri.NatGeoWorldMap") %>%
  addCircles(lng = ~longitude, lat = ~latitude,radius = ~sqrt(review_count))  %>%
  addCircleMarkers(data=fivestar,col="blue",group="fivestar") %>%
  addCircleMarkers(data=onestar,color='yellow',group="onestar") %>%
    #Layers control
  addLayersControl(overlayGroups = c("fivestar","onestar"),
                   options = layersControlOptions(collapsed = FALSE)
                   ) %>%
  # controls
  setView(lng=center_lon, lat=center_lat,zoom = 13)
map

Top five stars bars are more centralized and concentrated; while the worst one star bars are relatively more sparse in their location, and are not close to transportation hub.

Part 3: Social network analysis & regression models:

Social network of yelp elites:

library(readr)
library(networkD3)
library(visNetwork)
library(ggplot2)
library(plotly)
library(sjPlot)
  1. Network picture of top elite users

Among the elite users who have reviewed any of bars in Vegas, what are their relationship? Is there some inside network between these celebrities in Yelp? Or do they just review and travel separately? To figure it out, I find those elite users who also has at least one elite user friend and construct a network between them.

The picture beloew is generated from the top elite users of Yelp, which means these users’ fans >= 100, and the useful value of their reviews >= 900. In the picture, the marker only represents the user_id of these users. So it isn’t necessary to stick on the labels. The whole structure is the key.

## Every user has reviewed one of the bars in Vegas

name <- user[, c("user_id", "friends")]


net <- separate_rows(name, friends, sep = ",", convert = TRUE)
## No outside friends now
net <- net %>%
  filter(friends %in% user_id)
## Delete users with no inside followers

net <- rename(net, source = user_id)
net <- rename(net, target = friends)

netnode <- user %>%
  filter(user_id %in% net$source)


net <- as.data.frame(net)
p1 <- simpleNetwork(net, 
              nodeColour = "blue", 
              zoom=T,
              fontSize = 16)

##  There is no obvious network between the elite users
p1

As a result, most of the links are individual and not connected to another links. There exists A-B-C chains, which means A is B’s friend, and B is C’s friend, but no classical social network “circle” or group. On the other word, A is not C’s friend. Thus, as Yelp is not a social media, but a review app, elite users don’t interact with each other in the app. It is more possible that every elite user is the center of his/her fans, not the center of his/her peers.

  1. Network of bars in Las Vegas

While there is no group among the elite(influential) users, the other problem is whether this kind of group exists among the bars in Vegas. We assume that when a consumer goes to bar A, he/she will be more accessible to bar B in the same group because of recommendation or similarity. If there is a large portion of consumers go to both Bar A and Bar B, these two bars will be more possible to be in the same group compared to Bar C outside. Thus, we check all the reviews of top50 bar in Vegas based on review_count. Then we extract the reviewers’ id of every bar. After that, we define at least more than 50 common reviewers between two bars can be seen as a link. Finally, a network of bars is constructed below.

The links’ sizes mean the number of common consumers. The nodes’ sizes mean the number of reviews. While the colors of nodes mean different neighborhood in Vegas.

Double click the bubble, and you can see the neighborhood of this bar, and its most similar bar and this bar’s information. Attention: Some bars don’t update their neighborhood, so this part of these bars are blank. Feel free to find your next bar or next competitor.

rb <- barreview %>%
  group_by(business_id) %>%
  summarize(count = n()) %>%
  arrange(desc(count)) %>%
  head(50)
r50 <- barreview %>%
   filter(business_id %in% rb$business_id) %>%
   select(user_id, business_id)

## One bar have a branch in very close location also listed in top50, another bar also have a branch. We rename one of the branch to distinguish them.

r50 <- left_join(r50, bar, by = "business_id")
r50 <- r50[, c("business_id", "user_id", "name", "neighborhood", "review_count", "address")]
r50$name <- gsub('["]',"", r50$name)
r50$address <- gsub('["]',"", r50$address)
r50$name <- ifelse(r50$name == "Bachi Burger" & r50$neighborhood == "Southeast",  "Bachi Burger(SE)", r50$name)
r50$name <- ifelse(r50$name == "Firefly" & r50$business_id == "kRgAf6j2y1eR0wOFdzFAuw",  "Firefly(2)", r50$name)
r50 <- mutate(r50, barnum = as.factor(name))
barlink <- NULL
 for (i in 1 : 50) {
   for (j in i : 50) {
      a <- r50 %>% filter(as.numeric(barnum) == i)
      a <- a$user_id
      b <- r50 %>% filter(as.numeric(barnum) == j)
      b <- b$user_id
      c <- intersect(a, b)
      d <- length(c)
      tmp <- c(i, j, d)
      barlink <- rbind(barlink, tmp)
   }
}
barlink <- as.data.frame(barlink)
barlink <- barlink %>% 
  filter(V1 != V2) %>% 
  filter(V3 > 0)
barlink <- rename(barlink, source = V1, target = V2, value = V3)
barlink2 <- barlink %>% filter(value >= 50)
barlink2$value <- (barlink2$value / 25) ^ 2
barlink2$source = barlink2$source - 1
barlink2$target = barlink2$target - 1  
barnode <- r50[, c("barnum", "neighborhood", "review_count", "name", "address")]      %>% 
     distinct()
barnode$neighborhood <- as.factor(barnode$neighborhood)
barnode$rcsize <- (barnode$review_count / 100) ^ 2
mostsimilar <- NULL
for (i in 1 : 50) {
      a <- barlink %>% filter(source == i) %>% slice_max(value, n = 1)
      b <- a[1, 2]
      c <- barnode %>% filter(as.numeric(barnum) == b)
      d <- c$name
      e <- c$address
      f <- as.character(c$neighborhood)
      tmp <- c(i, d, e, f)
      mostsimilar <- rbind(mostsimilar, tmp)
}
mostsimilar <- as.data.frame(mostsimilar)
mostsimilar <- rename(mostsimilar, barnum = V1,  mostsimilar = V2, msaddress = V3, msneib = V4)
barnode <- barnode[order(barnode$barnum), ]
barnode$mostsimilar <- mostsimilar$mostsimilar
barnode$msaddress <- mostsimilar$msaddress
barnode$msneib <- mostsimilar$msneib
script <- 'alert("Neighborhood:" + " " + d.neighborhood + "  " + 
                 "Address:" + " " + d.address  + "    " + 
                 "Most similar bar:" + " " + d.mostsimilar + "   " + 
                 "Address of Most similar bar:" + " " + d.msaddress + "    " + 
                 "Neighborhood of Most similar bar:" + " " + d.msneib)'
p8 <- forceNetwork(Links = barlink2, 
             Nodes = barnode, 
             Source = "source",
             Target = "target", 
             Value = "value", 
             NodeID = "barnum",
             Nodesize = "rcsize",
             Group = "neighborhood",
             linkDistance = 300,
             fontSize = 30,
             opacity = 0.8, zoom = TRUE, clickAction = script)
p8$x$nodes$neighborhood <- barnode$neighborhood
p8$x$nodes$mostsimilar <- barnode$mostsimilar
p8$x$nodes$address <- barnode$address
p8$x$nodes$msaddress <- barnode$msaddress
p8$x$nodes$msneib <- barnode$msneib
p8

In this graph, we can find that there are so many links between these bars. Basically, all the top bars are clustered and not far away from each other. Of course, if you are a consumer seeking a different or similar bar compared to the previous one, you can just use this graph to find your preferred next bar. As an owner of a bar, you can also find your friend bars(or actually competitors).

Furthermore, it is obvious that the location doesn’t separate the bars in different groups. Perhaps Vegas is not a big place, so the distance doesn’t matter.

  1. Which neighborhood has most bars?

This picture tells you the distribution of bars in Vegas’ neighborhood. The strip is leading, and Eastside and Downtown are also great. However, if we take the average review scores into account, maybe Downtown is the best area for entertainment. Avgstar is the average score of bars in certain neighborhood, which can tell you the quality of the bars in certain area.

neighborhood <- bar %>%
  group_by(neighborhood) %>%
  summarize(avgstar = mean(stars), count = n())
neighborhood[1, 1] <- "Not known"
neighborhood$avgstar <- round(neighborhood$avgstar, 2)
neighborhoodgraph <- ggplot(neighborhood, aes(x = count, y = reorder(neighborhood, count), fill = avgstar)) + 
  geom_bar(stat="identity", width=1, color="white") +
  labs(x="number of bars", y="name of neighborhood", 
       title = "Which neighborhood has most bars?")

p3 <- ggplotly(neighborhoodgraph)
p3
  1. The gap between topbars and non top bars

This table tells you if we define the bars who have in average more than 4 stars and greater than 100 reviews are the top bars, how many of all the bars in Vegas are the top bars.

topstar <- mutate(bar, topstar = ifelse(stars >= 4 & review_count >= 100, "Top bars", "Non top bars"))
toppct <- topstar %>%
  group_by(topstar) %>%
  summarize(count = n(), avgreview = mean(review_count))


p4 <- tab_df(toppct, title = "The gap between topbars and non top bars")
p4
The gap between topbars and non top bars
topstar count avgreview
Non top bars 1139 106.49
Top bars 240 529.58

The result is that only 240 bars are top, and the rest 1139 are normal. The gap of average review counts in the two types of bars is huge. Even in Vegas, good bars are rare. But their popularity and quality are undoubtful.

  1. Which neighborhood has most top bars?
topstar <- topstar %>%
  filter(topstar == "Top bars") %>%
  group_by(neighborhood) %>%
  summarize(count = n(), avgstar = mean(stars))
topstar[1, 1] <- "Not known"
topstar$avgstar <- round(topstar$avgstar, 2)
topbargraph <- ggplot(topstar, aes(x = count, y = reorder(neighborhood, count), fill = avgstar)) + 
  geom_bar(stat="identity", width=1, color="white") +
  labs(x="number of top bars", y="name of neighborhood", 
       title = "Which neighborhood has most top bars?")

p5 <- ggplotly(topbargraph)
p5

In the distribution of top bars, we find Spring Valley and China town are great areas. They are easy to be ignored if we just concentrate to the location crowded with bars.

  1. What influences the elite users’ review scores most who has reviewed in Las Vegas.

Then let us dig deeper to discover the relationship among the attributes of elite users who has reviewed in Vegas’ bars. It seems that you will be more picky when you experience more places. But you may also be more conservative to criticize when your number of fans increases. The first linear regression graph shows this trend.

Move your mouse along with the line in the picture below, it will tell you when the elite users have certain number of fans, how much their review scores will be on average.
  average_stars
Predictors Estimates
(Intercept) 3.895950725 ***
review count -0.000084296 ***
fans 0.000152468 ***
useful 0.000000155
compliment hot -0.000002032
Observations 723
R2 / R2 adjusted 0.041 / 0.036
  • p<0.05   ** p<0.01   *** p<0.001
  1. Does the amount of review have a positive correlation with the bars’ reputation in Las Vegas?

It is said that some popular bars or restaurants may not worth a high review score. But at least this linear regression proves that bars in Vegas with more reviews tend to have higher reputation. They are worthy of their names.

Move your mouse along with the line in the picture below, it will tell you when the bars have certain number of reviews, how much their review scores will be on average.

lm2 <- lm(stars ~ review_count, bar, na.action = na.omit)
tab_model(lm2, digits = 9, show.ci = FALSE, p.style = "stars")
  stars
Predictors Estimates
(Intercept) 3.612174365 ***
review count 0.000136123 **
Observations 1379
R2 / R2 adjusted 0.005 / 0.004
  • p<0.05   ** p<0.01   *** p<0.001 
barstars <- ggplot(data = bar, aes(x = review_count, y = stars)) +
  stat_smooth(method = "lm", col = "blue") + 
  labs(x = "Number of reviews", y = "Average stars", 
  title = "The amount of reviews has a positive correlation with the bars' reputation in Las Vegas")
p7 <- ggplotly(barstars)
p7